Navigation

Showing posts with label resolution. Show all posts
Showing posts with label resolution. Show all posts

July 24, 2012

The iCamera (Nokia 808 Pureview) Part II: In the studio ;)

A still image taken in the studio by the Nokia 808 Pureview mobile phone
-- Please, click onto the image to access the uncropped full 38.4 megapixel image --


First of all, I'd like to clarify that this article isn't meant seriously. The Nokia 808 Pureview (808PV) isn't a studio camera. Medium format cameras and the Nikon D800E play in their own league when it comes to work in the studio. And they better do as the 808PV is positioned as a mobile phone ...

But just for the sake of having some fun, we can ask the question of how close the Nokia 808 Pureview (808PV) is actually being able to get to true studio cameras. And so we did :)

Above, you see "Young Fruits" in the studio, shot by the 808PV. I ordered some other young fruit but without going into to much detail, "Plan B" turned out to be it ;) You can click the above image to access the full resolution image. It isn't the JPG out of camera. It is the JPG after doing some post processing in Lightroom (no, the 808PV doesn't write RAW but this isn't a big problem for Lightroom). I basically improved contrast and adjusted colors a bit. I personally consider such things arbitrary as you can change settings in camera too. The image's EXIF data reads: 8.016 mm f/2.4, 1/100s, ISO 50. The image's physical size is 7.5 x 10.0 mm^2 (crop 3.45), i.e., the 35mm-equivalent EXIF data reads: 27.7 mm f/8.3, 1/100s, ISO 600.
It isn't actually simple to use the 808PV in the studio. This is the procedure I was going to follow:
  1. To save you a few hours of frustration:
    Do not use CameraPro, use the 808PV native camera app!
  2. Set the 808PV into creative mode, full resolution with 4:3 aspect ratio, super fine JPG, -1EV exposure compensation, sunlight white balance, ISO 50.
  3. The tricky part: Set flash to "Red eye reduction"
  4. Configure your studio flashes into slave mode.
  5. Focus with the Xenon flash allowed to act as focus assist, or with an external focus assist light source.
  6. Fire with 808PV Xenon flash:
    - either covered by paper to remove its effect from the photograph
    - or from a few meters distance where it appears darker
    - or alternatively enforce the ND filter which makes the Xenon flash appear darker too.
    The studio must be dark enough to make the 808PV use the flash at all. Because the "Red eye reduction" setting doesn't enforce flash in turn. The paper cover must not be too dark to hide it to the slave flashes though.
  7. Another tricky part: Your studio flashes need to be powered for f/2.4 ISO 50 (which is low power) or 3 stops higher power when using the ND filter (which is a bit above normal power, as would be used by ISO 100 f/9.6). But at the same time, the power must be low enough for your studio flashes to discharge twice within a fraction of a second. In my case, this required the flashes to stay at 1/2 power or below. Which I am perfectly happy with.
Following this procedure, the 808PV will fire the flash three times: a preflash to determine exposure, a first flash to make the eyes close and soon after, the final flash for the exposure which will have studio flash support again. The second flash will come too soon after the preflash to trigger the slaves. Some slave flashes can be configured with a trigger count. You can then use "2" and don't need the "Red eye reduction" setting. CameraPro seems to do the Red eye flash manually which implies that a second preflash is fired and synchronization fails. I had no problems to get the 808V to use anything between 1/100s and 1s exposure time. If you obtain faster than 1/100s or 1/200s exposure times though, you can dial in less than -1EV exposure compensation.

The 808PV has an ND8 filter which you can see with your naked eye when looking at the lens: there first is a protective glass, then a motorized protection shutter, then a motorized ND8 filter glass, and eventually there is the fixed aperture lens with focus motor (you can see the lens move forth and back). I wanted to make sure that the ND filter doesn't negatively effect image quality, so the image above has been made without ND filter. However, I have seen little or no negative impact of the ND filter on image quality. So, the above procedure may be easier to follow with enforced ND filter and I recommend to do so (if your studio flashes recharge quickly enough).


Result:

The results are pleasing. I invite you to click the image above to see the full size version. The quality is certainly good enough for most applications. And probably beats anything APS-C or FourThirds.

However, it isn't without quirks. First, enforcing ISO 50 is crucial. The pixels are small (1.4 µm) and therefore, the 35-mm equivalent sensitivity is ISO 600. I.e., the pixels aren't entirely noise-free and it doesn't help that they need some sharpening to get rid of diffraction blur. Moreover, the autofocus isn't fast and it isn't very reliable in studio modeling light (using the ND filter doesn't help either -- it is not switched off during AF ... ;) ). Therefore, the kind of studio work which would also use a tripod is quite feasible. But shooting is nowhere near as fast or reliable as with a dSLR.

Also, I had to ramp up the contrast a bit. It may be that the studio light from the sides create a little bit of stray light lowering contrast. Nothing serious though. And to be fair, the 808PV uses no lens hood :)

Look at the strawberry right to the peach, the bell pepper tip to its left, or the peach itself: IMHO, the level of detail is great.

However, before enthusiasm rises too high, I post the same subject shot with the Nikon D800E, for comparison:

A still image taken in the studio by the Nikon D800E 35mm digital SLR
-- Please, click onto the image to access the uncropped full 36.2 megapixel image --
While the level of detail in the image from the 808PV was great, it is simply stunning from the D800E. Just look at the peach'es hairs ...

So, while the 808PV delivers great results for static subjects in the studio which is good enough for professional use, it doesn't actually compete against the D800E or medium format. At least ;)

April 7, 2012

LumoLabs: Nikon D800 video function demystified

Nikon D800 FX mode 1080p video frame (click for original size)
The Nikon D800 full frame SLR camera has created a lot of buzz recently. Some would call it hype. While it is clear that its 36 MP still resolution is pretty much unparalled in the 35mm camera class, the final verdict about its video subsystem is still out. Esp. in comparison with Canon's 5DmkIII.

One point of interest has been how either camera actually creates its video frames. I now had a chance to apply LumoLabs' testing methology to a loaner D800 camera and figure it out for 1080p video in FX mode. I am having a look at live view performance too.

You may jump to the conclusion at the end if you just want to read what we found, igoring how we did it :)


Nikon D800 FX mode FullHD 1080p video

The title image shows one frame from a 1080p video taken with the Nikon D800 (in FX mode, it supports a number of crop video modes too). It shows a zone plate test chart which can be used to perform a sampling error frequency analysis.

Please, read falklumo.blogspot.de/2009/10/lumolab-welcome-and-testing-methodology.html to learn more about the testing methodology incl. access to the original of the test chart allowing everybody to replicate my analysis.

There is a bit of (gray colored) moiré from the printing process. This is because scaling and printing of zone plates is a non-trivial art in itself ;) You can actually measure the printer's native resolution by inspecting the printed zone plate chart. Below, you find a photograph of the print (in 14.6 MP resolution) allowing you to determine what moiré patterns are from the printing process actually.

Printed zone plate chart (still shot with a 14.6 MP camera, for reference)

However, all colorful moiré patterns are artefacts introduced by the D800 video system. It allows us to precisely measure how it works. Let's have a close look at the one of the two center discs:

Analyzed region of interest in the D800 video frame

The big discs are constructed such that the 1080p Nyquist frequency emerges at its outer circle. The two center discs have their edge at twice this Nyquist frequency and the four tiny discs at four times this frequency. Therefore, the false color moiré disc emerges at (149px/258px x2) or 1.155x the 1080p Nyquist frequency (1247 px). This means that the Nikon D800 samples ~1247 horizontal lines from its sensor.

Now, let's make a back-of-the envelope calculation:

An FX frame in video mode is taken from a 6720 x 3780 px region (which actually is a 1.095x crop from the full 7360 x 4912 px frame (this information is from the Nikon user guide, translating physical dimensions into pixels). Because 3780 / 1247 = 3.03 and because 1% is our measurement error, we have proof that the Nikon D800 samples every third horizontal line from its sensor.

A second result is that the ever so slightly color moiré for horizontal frequencies disappears at the Nyquist frequency. The D800's AA filter is effective here, the remaining moiré is from the printing. The D800E would have a bit of additional color moiré here, but by far not as strong as in the vertical direction. So, I believe that the Nikon D800 samples every vertical row from its sensor.

Below is what I believe how Nikon implemented line skipping:

Likely D800 sensel sampling matrix

and here is a slightly more symmetrical scheme which I cannot entirely exclude although I think it isn't used in this mode:
Unlikely sensel sampling matrix
If you look at the likely sensel sampling matrix, you'll see that all sensels which are read out (the ones with a color) result in a new RGGB Bayer matrix of sensels. Which has the advantage that a standard demosaicing algorithm is applicable to create an RGB frame.

This is similiar to what the Canon 5DmkII did actually. However, there is one important aspect where the D800 is different:

A native 1080p video frame is 6720 x 1260 px, demosaiced to a 2240 x 1260 px RGB frame.

And the final 1080p video frame is further downsampled 7:6 to 1960 x 1080 px which gives the D800 a slight edge in resolution and edge flicker behaviour over a 5DmkII.


High ISO noise in video

What we found has one important consequence: High ISO noise in video! Because of the FX video crop and skipping two thirds of sensels, the ISO performance in video is shifted by a factor 3.60. E.g., At ISO 12,800, the noise looks (as bad) as at ISO 46,000 from a camera using all available sensors for video (except for the 16:9 ratio crop of course).

You may note however, that the D800 still samples 6720 x 1260 sensels for a 1920 x 1080 frame or 4.08 sensels per pixel. For this reason, at ISO 12,800, the noise looks (as good) as at ISO 3,200 from a still image when pixel peeping at a 100% (1:1) level. So, pixel noise in D800 video is 2 stops less compared to still while it could have been 3.85 stops less when reading out a maximum of sensels. If you consider this bad or good is up to you.

Below, I have extracted frames from the ISO comparison performed by crisislab.com:

Video noise comparison D800 vs. 5DmkIII -- original frames (c) 2012 crisislab.com
On the left hand stripe, I have shifted the D800 samples two stips down and I think, it is a good match for the 5DmkIII performance then.

From that, I can already conclude that the 5DmkIII reads out all its sensels, i.e., does no line skipping. However, I didn't run a resolution analysis for the 5DmkIII. However, hearing about resolution complaints for 5DmkIII video, I think they bin pixels before read out. This improves noise and aliasing performance but unlike downsampling, doesn't help the resolution.


Nikon D800 Live View implementation notes

I have applied our testing methodology to Nikon's live view implementation too.

D800 live view, photograph of the rear LCD (no zoom level)
You see the same false color moiré discs which we have analyzed already. Of course, there is some strong additional moiré from the LCD rasterization. I.e., the D800 only reads every third line when activating live view (in the example, it is FX video live view).

If we zoom in, we get a result as follows.

D800 live view, photograph of the rear LCD (high zoom level)
You now different false color moiré disc, they have moved outwards. The sampling frequency is  (1692px/1935px x2) or 1.749x the 1080p Nyquist frequency (1889 px). Because 3780 / 1889 = 2.00, we have proof that the Nikon D800 samples every second horizontal line from its sensor when zooming enough in live view.

In live view, the D800 switches from third line to second line skipping when zooming in!

Lessons for manual focusing: (1) zoom in and (2) focus onto vertical structures which have twice the resolution in live view! Focus on trees, edges of buildings rather than horizon or roof top.


Conclusion

The D800 creates FX 1080p video in the following way:
  1. Crop a region of 6720 x 3780 sensels (crop factor 1.095).
  2. Read only every third line out of this region, but all sensels in a line. The result is an 6720 x 1260 sensel RGGB Bayer pattern which can be demosaiced.
  3. The resulting 2240 x 1260 RGB image is downsampled 7:6 to the final 1920 x 1080 px resolution.
  4. Compared to an optimum architecture, only 1/3.6 of sensels are read which makes the D800 loose up to 1.8 stops in high ISO video performance.
  5. When zooming into a live view image, the D800 switches line skipping from 3x to 2x.
  6. Manual forcus should use zoomed live view focusing vertical edges.
Overall, I am personally pleased with the implementation Nikon has chosen. It refines an idea originally used in the 5DmkII which is more difficult to implement due to the higher overall number of pixels. Because of downsampling from 1260p to 1080p, I actually expect slightly better resolution than from a 5DmkII or a camera which bins sensels prior to demosaicing.

On the other hand, there will be no more excuses for line skipping in the future. Not after Nokia got rid of it in their 41 MP 808 mobile phone ...


Enjoy your read :)
Falk

March 5, 2012

The iCamera (Nokia 808 Pureview) Part I





Why "The iCamera"?

Because in my opinion, this term would best describe what this camera is all about. Camera? Yes, I consider the Nokia 808 Pureview to be a camera (which happens to have an embedded phone), not the other way round. Nokia may not like this perspective. Not yet. Further down, I am going to explain why it is such a great revolutionary camera. But first, let me explain why I call it the iCamera. And no, I will not say anything about the iCamera's embedded phone function. Please, refer to Nokia for further details about the phone.

Remember the iPhone back in 2007? The year the 808 project was started? "It is an iPod, a phone, and an internet communications device". Some laughed at it, but it became the blueprint for almost every smartphone made since. And a big success. And what matters most: it came from a company who never did a phone before! Launched after years of hidden development.

Now, you have "a camera, a phone, and an internet communications device". Some laugh at it, but it may become the blueprint for almost every compact camera going to be made. It comes from a company who never did a camera before! Launched after years of hidden development.

Nokia may not realize what they did and actually may still screw it up. But if they play their cards right, they can create an entirely new market and rule it! A photography market. Just let's forget for a second that the 808 is also a phone. The iPhone is also an iPod. So what?

And so I call it the iCamera.


Why is it revolutionary?

A product becomes revolutionary if it combines existing technologies in a way that the result leapfrogs several generations of competing products and seems to contradict common wisdom. The 808 does this. Let me explain in detail now.

First, let me classify the iCamera. You need to read my paper about camera equivalence though. You'll find it here:
It explains how to correctly compare cameras when they have different sensor sizes. The 808 is two different cameras at the wideangle and the long tele end. And another different camera in video mode. I give three 35mm equivalent cameras at both ends now (ISO are the minimum equivalent values), followed by 5 other cameras serving as a reference:
  • Wide iCamera:      28mm/8 iso640 (38MP, 169g)
    exact: 38.4 MP (4:3), 27.7 mm, F8.3, ISO 690 (crop 3.45)
  • Long iCamera:      77mm/22 iso6400 (5MP, 169g)
    exact: 5.0 MP (4:3), 76.6 mm, F22.9, ISO 5300 (crop 9.55)
  • Wide video iCamera:      26mm/8 iso640 (HD, 169g)
    exact: 2.1 MP (16:9), 25.7mm, F7.7, ISO 590 (crop 3.2 wrt 4:3)
  • Apple iPhone 4:      29mm/22 iso4800 (5MP, 137g)
  • Wide Olympus E-P3 + 14-42/3.5-5.6II:      28mm/7 iso400 (12MP, 481g)
  • Long Olympus E-P3 + 14-42/3.5-5.6II:      84mm/11 iso400 (12MP, 481g)
  • Wide Nikon D800 + 24-70/2.8G:      24mm/2.8 iso50 (36MP, 1800g)
  • Long Nikon D800 + 24-70/2.8G:      70mm/2.8 iso50 (36MP, 1800g)
Please, refer to the Nokia White Paper for the 808 specification and further details.

Even though the tele effect (Long iCamera) is achieved via simple cropping, it is equivalent to a different camera. Very much like a zoom lens which yields different focal lengths and apertures at both ends. Equivalent means you should expect the same indistinguishable image quality from a full frame 35mm camera with the given specification. Please feel invited to read the paper above to understand the details (applies to Nokia employees too: please understand why your way to zoom via crop is not equivalent to a constant aperture zoom).

In a nutshell: the above list allows for a direct, quick and no-nonsense comparison of available products.

And it shows that the iCamera is physically very close to a mobile smart phone (actually, Nokia says it is one) but its optical performance is rather close to a mirrorless system camera, only significantly bettered by a big full frame SLR with a fast lens.

So, in optical performance, the iCamera leapfrogs all mobile phone and P&S cameras, while in resolution, it additionally leapfrogs all mirrorless system cameras and even APS-C SLRs.

Of course, in order to make this claim, we have to check if the 808's lens is up to the task. So, let's refer to some sample images which have been published already

Image taken with the Nokia 808 Pureview (the iCamera). © 2012 Nokia (click to access original)

The above image is taken with the Nokia 808 Pureview (the iCamera). The image is mildly post-processed by me in a way typical for other photo cameras in JPG mode (the image is a bit sharpened and noise-reduced) to provide an easier reference for a direct comparison (the 808 seems to apply no or almost no post-processing to its results which is a good thing for photo enthusiasts).

The above image is an example of good image quality for a mobile phone or almost acceptable image quality one would expect from a system camera with a kit zoom. Most details are resolved but there remain a few sharpening/denoising artefacts.

So far, this is nothing spectacular or worth talking about.

But what if I tell you that the above is an 18x crop (1:1 or 100% crop) taken from the following image?

Image taken with the Nokia 808 Pureview (the iCamera). © 2012 Nokia (click to access original)

The above is the full and same image as shown above! Only resized to the typical web size. And now and immediately, you may understand why the iCamera is like no other camera before!


The Zeiss lens

Moreover, this example clearly shows that the lens is up to the task and able to resolve the incredible resolution of the sensor. Closer inspection of the original images reveals even the corner resolution is  good even if it doesn't fully resolve the sensor, still beating most other available cameras in the corner. A more complete analysis was done by Werner Ruotsalainen: Nokia 808 resolution tests. He concludes that even the smaller 5 MP images from the 808 beat 10 MP images from a P&S in resolution! And there are reasons to believe this to be true.

But how can a mobile phone lens resolve such tiny detail? Good question. As I explain in my paper about camera equivalence, it becomes increasingly harder to resolve many pixels when shrinking the sensor size.

But famous lens maker Zeiss did an incredible job solving this problem: By glueing 5 lenses together into one group, all having aspherical surfaces and one using ED glass, they created a lens with unprecedented optical performance and very tight manufacturing tolerances. The problem is larger than it appears: The lens aperture is only 3.34mm while the image circle 8mm further down the optical axis is a whopping 12.5mm, almost 4x the aperture. If the aperture were as large as the image circle, this would be an f/0.6 lens! So, the lens elements nearer to the sensor are larger than they appear when looking at the camera front. Such a lens must be mounted to the sensor assembly on extremely tight tolerances or it is decentered. We have to keep in mind that the iCamera's lens is diffraction-limited at F2.4! Nokia uses a live manufacturing method where live view from the camera is used to fine calibrate the lens when assembling the sealed camera module. That's fairly innovative by itself and definitely virgin territory. This method won't work for an interchangeable lens camera. More on this below.

UPDATE 2012, March 5:
Today more information about the lens was made available by Nokia and Zeiss.
Zeiss 808 8mm/2.4 lens
source: http://conversations.nokia.com/2012/03/05/nokia-808-pureview-carl-zeiss-science-of-making-the-perfect-lens/
The lens is depicted in the image above and as it said in the source, it is made from a special plastics rather than glass. This is only possible for small lenses like the one in the 808 but has the advantage that much more complex surfaces can be made to precision. That seems to be part of the recipe how to achieve the high optical performance despite the small footprint. END OF UPDATE.

So, we conclude that the optical performance figures are hard to believe but they seem to be real.

But this isn't everything yet. We saw that the wide iCamera is equivalent to a camera with rather low ISO setting. So, it should have good dynamic range. And indeed, this seems to be true. I treated the above image in an HDR manner and came up with the following result:

Image taken with the Nokia 808 Pureview (the iCamera). © 2012 Nokia (click to access original)

The above is is still the same image as shown above! But now we used the iCamera's dynamic range to heavily boost shadows to illuminate the black regions of the original image. That's not normally possible with images from mobile phones.

The reason why this works is that the iCamera has a large sensor, much larger than mobile phones or P&S cameras. It is almost as large as the sensor in the mirrorless Nikon 1 system cameras. The equivalent camera parameters express this as an equivalent ISO value of 640 which is known to be good enough to have enough headroom for extending an image's dynamic range (like I did above).

I hope that all the text above answered the question why the iCamera is indeed revolutionary.


One more thing ...

It wouldn't be The iCamera if there wouldn't be one more thing ... :)

All still photo cameras have the problem that the sensor has many more pixels than there are in HD video. But it is hard to read out all pixels of a still image (10 MP or more) 24, 25, 30 or even 60 times a second. Therefore, still cameras only read a small fraction of its pixels to make the video stream, known as subsampling or line skipping. The effect is a significant degradation of image quality in video mode: there is noise, line flicker, color moiré and the result is no match for HD content produced with so called 4k cameras or cameras with supersampling such as the Canon C300. Such cameras cost $15,000 or more (a notworthy exception is the Panasonic GH2 though which made it the camera of choice for serious video work on a budget).

And what shall I say? The iCamera does it too, not supersampling 8MP (C300) or 16MP (GH2) but supersampling all 33,593,616 pixels (16:9) 30 times every second! That's one billion pixels the iCamera processes every second. In a mobile phone. This is crazy!

As a consequence, the wide iCamera could have the same good low light capabilities in video mode as the legendary Canon 5DmkII which does line skipping, but without the Moiré and line flicker problems.

And because no graphics processor (GPU) obviously can handle this data rate (otherwise, HDSLRs would be able to do it too), Nokia designed a special chip (the scaling processor) which sits in the camera module between the sensor and the GPU. And Toshiba managed to make a sensor which can output the equivalent of 8 GBit/s. So, there is a large CMOS sensor, a scaling processor, a graphics processor and the ARM CPU chewing each others output to get the job done.


The people behind

Nokia was kind enough to emphasize the role of three persons in this project:

Eero Salmelin and Juha Alakarhu (Nokia "pureview inventors")
Source: http://conversations.nokia.com/2012/02/29/zooming-in-on-nokia-pureview/
Damian Dinning (Nokia "imaging guru")
Source: http://conversations.nokia.com/2012/03/01/nokia-pureview-qa-with-damian-dinning

I believe that the creation of the iCamera was driven by opportunity rather than strategic planning. Nokia decided to give engineering green light to pursue the project and see where it leads to. And I guess they are still watching. The Nokia press conference only "mentioned" the 808. And that's the real difference between the iCamera and the iPhone: the iCamera isn't top-level driven, it is not born out of vision from the leaders. It is born out of engineering vision and often, this is not enough.

So, here is my humble advice to Nokia managers who are only watching the 808:


Implications

The iCamera is such a groundbreaking device that Nokia must no longer consider themselves a phone maker.

The iCamera is such a groundbreaking device that Nokia must no longer consider themselves a phone maker.

( I said it twice on purpose.) Like the iPhone made Apple drop the "Computer" in "Apple Computer", Nokia may consider to drop the "Phone" in their brain. They may continue building great phones just like Apple continues to build great computers. But now, they have to become a camera maker and whatever this leads to. Nokia engineers may have understood it already: They provide a tripod mount and a decent camera user interface.

It is useful to consider what the iCamera could be in another form factor (less phone like) and to remember that interchangeable lenses wouldn't necessarily deliver (cf. above). The "lens module" concept then comes to mind. Where a lens module is exactly what the camera module is in the 808: A sealed module containing lens and sensor. And therefore, a system camera would become:

A mirrorless interchangeable module camera (MIMC).

Nokia wouldn't be first (Pentax Ricoh Imaging with their GXR line is first). But an 808-based MIMC would be the first camera where this concept actually makes sense: Because it delivers a level of performance not possible otherwise. And at an attractive price point considering the Nokia 808 minus the phone isn't more expensive than a system camera's lens.

Nokia could create the future and dominant kind of camera market between the phone and 35mm full frame system cameras. And they could reestablish a European camera industry just like Apple revitalized a dying US phone industry. There are good cooperation partners to the task, like Zeiss or Leica. Nokia may even consider to make a GXR module to test the waters.

Anyway, personally I plan to release more parts to this article after I got the chance to run some laboratory tests. It will be exciting to see what all of this leads to.

Enjoy your iCamera.

February 21, 2012

LumoLabs: Camera equivalence

Various parameters, or variables of a real camera or a reference camera are depicted above

In preparation of an article discussing the advantages and disadvantages of various sensor sizes for a given camera performance, I try to set a common ground for such discussions.

I have prepared a white paper which dives much deeper into the topic than is possible in this short blog article. You may find it here:
The short version is this: An image contains no information whatsoever about the size of the sensor within the camera which was used to capture it. None. Nothing. Nada. (except EXIF of course ;) ) The proof is beyond the scope of this blog article and the article only gives some clues. But this is a fact, trust me.

Therefore, all cameras which could have captured a given image create a so-called equivalence class: they are all equivalent, producing indistinguishable images. And they have different sized sensors! By camera, I mean a camera with all the parameters defined it used to capture an image, such as the variables shown in the title image. Changing any variable "creates" a different camera. The exposure time used to capture an image is defined implicitely too: the one giving correct exposure (and it is a constant of course for indistinguishable images).

The following image shows an equivalent camera where the sensor has only half the size of the first or reference camera, i.e., an equivalent crop-2 camera:
The camera's lens has the same absolute diameter but it's focal length is shorter to maintain a common field of view. The equivalent crop-2 camera has a different F-stop and ISO sensitivity.


Main claim:

Any discussion about the impact of varying sensor sizes must be based on cameras made equivalent first. Otherwise, any comparison will just reveal the inequivalence of parameters the respective cameras have been set to and nothing else. And such a result would be trivial, known and not worth a further discussion.

Such trivial results are that a larger sensor produces a more shallow depth of field or less image noise. This is not true! Because it just means that the cameras were used with non-equivalent settings, e.g., with lenses of different diameter d which means with lenses of different weight and cost. Another example are ISO comparisons between cameras with different sized sensors but ISO kept the same. Such comparisons are pointless! Instead, compare a FourThirds camera at ISO 100 with a full frame camera at ISO 400 because only then they are equivalent. Not doing so just compares the size of lenses which a ruler can do just as well.


Secondary claim:

Once equivalent cameras are compared, results start to become interesting. Because now any deviation is due to deviations with respect to an ideal camera. Such like a lens with aberrations, production or design tolerances or compromises in a CMOS production process. The white paper explains that such deviations are generally expected to be larger with smaller sized sensors. Of course, one such deviation is obvious: when an equivalent camera doesn't exist for a sensor size, e.g., because an f/0.1 aperture is unfeasible.


I will follow up this article with a more complete article of the impact of sensor size on image quality.

Stay tuned and enjoy your read :)

April 26, 2011

A hypothetical Pentax DFA* 500mm F5.6 ED(IF) SDMii

Kestrel © 2011 Falk Lumo

  Kestrel in the wild near the garden.
 The equivalent focal length is 2600 mm. Shot with a Pentax K-5 and 1.7x AF converter using a 500mm lens. Cropped to half size.
Kestrel


Kestrel in the wild near our garden.
 The 35mm-equivalent focal length is 2600 mm. Shot with a Pentax K-5 and Pentax 1.7x AF converter using a 500mm lens. Cropped to half size.

The Pentax system as great as it is does still lack a super tele lens option from Pentax which is in production. This made me wonder which lens exactly Pentax should do to both fill the empty space and to attract new photographers into their system.

Pentax once was famous for optically great and affordable super tele lenses such as the Pentax FA* 600mm F4 or gorgeous 645 600mm F5.6 etc. It is a shame they take so long to revive their tradition of great long glass.

To my big surprise, the answer which lens to start with was pretty easy:
A Pentax DFA* 500mm F5.6 ED(IF) SDMii.

Before I'll dig into details, let me explain why. First, shorter lenses don't make sense with a popular DA* 300 F4 or DA* 60-250 F4. After all, digital cameras are crop machines and need larger steps between lens options to make sense. But an even faster or longer lens becomes too expensive to be a smooth enough upgrade (as detailed below, a 600/5.6 would have to be 80%, a 500/4.5 100% more expensive). Same goes for a very long zoom lens which would become too heavy and too expensive too. Moreover, f/5.6 is really fast enough with the advent of great sensors such as the one in the Pentax K-5.

However, it must not be slower than f/5.6 (or maybe f/5.4) because then the Pentax AF system wouldn't work anymore in conjunction with an 1.4x tele converter. Yes, it does work with f/5.6 or f/7.8 effective. How do I know? Well, the above photo was shot with a Sigma 500mm f/4.5 APO manual lens and the Pentax 1.7x tele AF converter which is f/7.7 effective. The AF works flawlessly and fast! Heck, the K-5 even gives me a stabilized 850 mm autofocus lens this way :)

A 500/5.6 lens really is the sweet spot between long enough, fast enough, not too heavy and not too expensive. And it is the missing option from Sigma, Sony, Canon and Nikon! Their product suites are all so 2000-ish! Optimized to be fast, expensive and heavy rather than the versatile resolution super weapons modern sensors need so much.


Pentax DFA* 500mm F5.6 ED(IF) SDMii details

I compared a couple super tele primes from Sigma and Canon (Canon MSRP prices divided by 1.34 to meet Sigma MSRPs, e.g., for the 800/5.6). The following two formulas are the best description to predict weight and price:

Weight [kg] = D[mm]^2.20 /12000 (typical error is +/- 15%)

Price [MSRP in USD] = D[mm]^3.26 /1177 (typical error is +/- 25%, street price shouldn't be higher)

D is the diameter = focal length / f-stop number.

D^2 is the expected term describing how glass surface grows with diameter. 2.2 accounts for increasing thickness, I guess. Price should be proportional to weight but isn't. It has an extra factor D probably accounting for increasing rareness and hand-made production steps. Anyway, the above formula allows to describe the lens in fairly accurate terms:
  • Lens: Pentax DFA* 500mm F5.6 ED(IF) SDMii
  • Mount: Pentax KAF
  • Price: 1940 USD (MSRP)
  • Weight: 1630 g
  • Filter size (front): 91 mm
  • Length: 300 mm
  • Close distance: 4 m
  • Lens elements / groups: ~12 / ~11 (e.g., like the Pentax A* 645 600mm F5.6).
  • Stabilized: Yes (in body)
  • Autofocus: Yes (SDM, version 2 (ring motor); screw drive supported; focus limiter in firmware)
  • Teleconverter: Yes, optomized to support Pentax DFA 1.4x SDM converter (tba ;) )
  • Image circle: 43mm (full frame)
  • Weather-sealed: Yes
  • MTF characteristics: Blur widths below 2px in the center, similiar to DA*300, outresolving the sensor x2 ("made for cropping" (tm)).
Option:
  • Pentax DFA 1.4x SDM converter
    700 mm, F7.8, phase AF supported in daylight.
I am pretty sure that this lens would sell in a volume high enough to make it an economically viable option for Pentax. And drag new users into the Pentax system. And keep existing users happy as well ;)

Pentax, where is your pre-order page? :)

July 21, 2010

LumoLabs: Shutter-induced blur with an SLR camera



A recent observation made by us and others was that shake reduction efficiency for the Pentax K-7 camera seemed to have a weak spot around about 1/100s and less. Something nobody could really understand and not everybody was able to confirm.

Therefore, we decided to try to answer an old and fundamental question for SLR photography: To which extent does the mechanical focal plane shutter and the mirror slap negatively influence image sharpness? Especially in the digital age with its theoretically rather high image resolution. We, this means two friends (Henning and Rüdiger) and myself (Falk). And of course, we decided to focus our study to the Pentax K-7 SLR camera in order to provde an answer to the observation mentioned above.

The short story is that we managed to find the answers. All our findings are written down in detail in a LumoLabs White paper:

-> http://www.falklumo.com/lumolabs/articles/k7shutter/index.html.

Please refer to this document to actually understand the work we have done. In the following, we will summarize our findings without explaining how we got there. However, note that 4 different camera bodies, data from 4 testers, 8 lenses and two firmware versions have been used. More than thousand test shots and several thousand accurate blur data measurements have been aggregated. High speed video, acoustic recording and acceleration measurements complement the data. So, we assure that the result describe the general behaviour of a Pentax K-7 SLR camera. Pentax has obtained a copy of the paper to be used at their discretion.

We will make no statement about how the results relate to other SLR cameras. Except for a quantitative comparison with one Pentax K20D SLR camera.


Results:

  1. The mechanical focal plane shutter indirectly can increase the blur in an image. The exact amount of additional blur depends on the direction in the image. It is zero at a vertical contrast edge (aka yaw blur, blur due to yaw movement). And it is up to 11 µm (on average) at a horizontal contrast edge (aka nick blur, blur due to nick movement).

    The exact amount of average blur is shown in the opening figure of this article. It has its maximum for shutter speeds of about 1/100s to 1/80s. It is less than 5 µm for 1/25s and slower. Or 1/250s and faster.

    Note that any single image can be affected more or less. Add or subtract +/-50% to get an idea of variation from image to image.

    Note that one pixel is 5 µm large and the blur effect is only visible if all other sources of blur are very well under control (sharpening, defocus, shake, subject blur, lens abberation, noise etc.). Normally, these other sources mask the effect. Nevertheless, if you want tack sharp images then you need to understand the shutter blur effect.
  2. The effect for the Pentax K-7 is larger than for the Pentax K20D. About 2 - 3x larger.
  3. Mirror slap or shake reduction have no negative or positive impact on the effect. Shake reduction works as advertized but cannot counteract the perturbation from the focal plane shutter as it is too fast really. Mirror slap is very well dampened in the K-7 camera and has no negative impact on image resolution except on a weak tripod.

    There is a delay of about 10 ms between end of mirror slap and begin of shutter operation which suffices to keep the mirror slap perturbation out of the image.
  4. The blur effect is an indirect one:

    First, the moving masses of the shutter (curtain etc.) make the body move (with surprising speed and acceleration of its stiff body!).

    Second, the body movements cause a classical blur effect lasting as long as the shutter works. The K-7 shutter is faster and stronger than that of the K20D probably increasing the effect by some 60% or so.

    Third, the body accelerations cause additional vibrations in the imaging sensor which last a bit longer than the first shutter curtain operates and which magnify the effect by another 60% or so.

    Preventing the first from happening (which requires a heavy and sturdy tripod) will kill the effect. There is no "loose" magnetically held imaging sensor and no negative direct impact from shutter curtain or mirror slap causing air flow in the mirror box or whatever.
  5. In practice, you'll only see any effect with wide angle lenses.

    At about 1/100s you would normally have blur due to free-hand shake (we can ignore the case of a tripod as only weak tripods would cause any trouble with the shutter). At 50 mm and longer, the shutter blur will be masked and at 30 mm it will have comparable magnitude. It is at 10-20 mm that the effect will be noticeable most.

    In these cases, we highly recommend to shoot at 1/25s (or slower) and to enable shake reduction as it is highly efficient at such exposure speeds. The images will be sharper than at 1/100s!
  6. Early efficiency tests of the K-7 shake reduction suggested that it may be ineffective at fast shutter speeds as required for long focal lengths. This was a preliminary conclusion we proved to be wrong.

    The Pentax shake reduction is effective even at 1/500s! It just cannot prevent the shutter blur at about 1/100s. We may soon publish an update to our SR guide reflecting this.

So, here you have it in a nutshell. Please, refer to the full paper before asking questions. The paper is available as HTML and PDF (linked from the top of the paper). It is recommended to download and read the White Paper on "Understanding Image Sharpness" first.

UPDATE (2010 July, 28):

We checked if the new firmware release 1.10.00.25 released earlier today brought an improvement. The answer is NO.

We've run a number of measurements and within the limits of our very good measurement accuracy (about 0.10 to 0.15 pixels error margins) we cannot see an improvement.

[end of update]

Frequently Asked Questions:

Q: Did you study the K-x, do you know if it has a similar effect?
A: No. But anybody is invited to replicate our study for the Penatx K-x :)

Q: Is the shutter blur in the Pentax K-7 a defect?
A: No, any SLR shutter for any make causes blur to some degree. We just wished for the Pentax K-7 that it would be as small as it is for the K20D. We publish this partly to remind all camera makers that we watch their work ;)

Q: Does switching off shake reduction lead to sharper images?
A: No.

Q: Does mirror lookup work around shutter blur?
A: No.

Q: Does a tripod work around shutter blur?
A: Sometimes. If it is rock solid. A normal tripod most likely won't help much.

Q: Why does a longer exposure time work around shutter blur?
A: Because during the majority of the exposure, the shutter won't move and what you get is an average blur.

Q: Why does a shorter exposure time work around shutter blur?
A: Partly, because there simply is less time for anything to blur. Partly, because stimulated vibrations cause no harm after the shutter already closed.

Q: May I ask questions without reading the White paper?
A: No.

Q: But I don't understand the White paper!
A: How do you know without reading it? ;)

Q: Will you win a Nobel price for this crazy shit of work?
A: No. Alfred Nobel forgot photographers ;)

Further reading:

Enjoy the read ;)

June 17, 2010

LumoLabs: Understanding Image Sharpness

We have prepared our first White Paper. It shall serve as a base to better understand our methodologies to measure image resolution and more importantly, it is meant to help understand what factors can prevent an image from turning out tac-sharp.


LumoLabs website

We changed the layout of our site too.

LumoLabs is now at www.falklumo.com/lumolabs and hosts a repository of articles.

Therefore, we will use the blog to announce new articles or important updates to followers and interested parties. And to enable their discussion.

The actual articles are not posted as a blog article as its format was deemed unsuitable. But you'll find links to both the online article and a printable PDF version. If possible, we always recommend to download and read the PDF version. The PDF version does update more frequently too ;)


Understanding Image Sharpness


(Sample chart form the article)
Hint: The article image URLs actually open as larger images as they appear embedded in the article.


This article is a recommended read for anybody loving to dig into technology and who isn't afraid of a bit of math.

It's abstract and table of contents is:

Abstract
This White Paper is one in a series of articles discussing various aspects in obtaining sharp photographs such as obtaining sharp focus, avoiding shake and motion blur, possible lens resolution etc. This paper tries to provide a common basis for a quantitative discussion of these aspects.

Table of Content
1. Measures
1.1. Modular Transfer Function
1.2. Blur
1.2.1. The hard pixel
1.2.2. The perfect pixel
1.2.3. The real pixel, sharp and soft
1.3. More realistic resolution measures
1.4. Combining blur
2. Sources of blur
2.1. Defocus
2.1.1. Ability of deconvolution operators to reduce defocus blur
2.2. Bayer matrix and anti aliasing
2.3. Diffraction
2.4. Lens aberrations
2.4.1. Defocus, Spherical aberration, Coma, Astigmatism
2.5. Shake
2.5.1. Measuring shake
2.5.2. Expected shake
2.5.3. Empirical results
2.5.4. Tripod classification
2.6. Motion blur
2.7. Noise
2.8. Atmospheric perturbations
2.9. Precision and calibration
3. Practical considerations and examples


Please, proceed here:


 

June 19, 2009

Comparative resolution study K-7 vs. K20D


This blog article was initially meant to be part of the noise article. However, the matter turned out to be much more intricate than I thought (and somehow, it still is). So, I made it a separate article again owned by falconeye.

I would have loved to make it a short answer. However, I am going to tell you a story instead. I am sorry that Falk Lumo will have to wait with his catwalk beauties ;)

In the top of the article, you see a (synthetic -- more on this later) resolution test shot for the K-7. Please, click on it and open it in original size, and scroll to make the center visible. Note that what you see is the inner part of a special edition of the ISO 12233 test chart. Special because it contains insets of 4x resolution. I made this edition to be able to measure today's dSLRs. It is printed in A2 size and the printer's dots start to form Moiré at figure "8.5" in the chart insets, i.e., between "8" and "9". No big deal, just something to keep in mind.

For more information about the ISO 12233 test chart, you may want to visit this link:
http://www.imatest.com/docs/sfr_instructions.html

Also, note that the test chart only fills 86.7% of the picture height. As a consequence, the chart figures have to be multiplied by 460 (rather than 100) to yield LW/PH (line widths / picture height). Read, the K-7's Niquist limit is at chart figure "6.7". There shouldn't be any structure beyond this number.

Now, look at fields labelled "7" or "8" (we ignore "9" and higher for the reason given above). There still is structure but it is just Moiré, not the proper line pattern which is printed. So the lens plus AA (anti-alias) filter outresolve the sensor and we pick up a pseudo pattern which basically is a sampling error (which is what is called a Moiré pattern). Would lens plus AA filter be at the sensor resolution or below, than we would have uniform gray fields labelled "7" or "8".


First result:

The K-7 has an AA filter which allows to outresolve the sensor. It better has :)


Now comes the tricky part. Because we want to compare with the K20D which we must assume does have an AA filter which allows to outresolve the sensor, too. How on earth shall we measure a resolution beyond the sensors' capabilities? If I a had the accuracy and software of a testing lab at my disposal, I would extract these numbers from subtle details in the respective MTF figures; but I don't.

First, let's verify that the K20D is close indeed. Here is the corresponding synthetic resolution test shot for the K20D:



Oh yeah, just as nice :)
(About 3 visible fine white rings in the center circle.)

Ok, let me get serious now (which means that most likely, you're gonna run away before the end of the article -- just make sure to join in for the verdict ;) ). In order to study the AA filter's effects, I will apply a technique called superresolution. This technique involves taking multiple test shots (16 in this case) and creating a synthetic image at twice the native resolution, i.e., an 58 MPixel image! In order to construct this image, a "kernel" (deconvolution with a native pixel's PSF (point spread function)) is involved. By pure accident, I happen to have a kernel which rendered me good service with my K20D. Let me call this kernel "Falk1". I refrain from describing the exact procedure here... Let's just accept that, by some miracle, we now have test images at twice the resolution.

Because they are so large, I will only present crops (200% crops = 100% crops of superresolved image). Let's call them supercrops. First, for the K-7, next for the K20D.




Now, we see 5 rings and we suddenly see a difference, too! However, let's be cautious. The K-7 image has some artifacts. In the K-7 supercrop, the lines have more false color and more interestingly, the end of lines have a bubble artifact. E.g. look at the "5". From my experience with the superresolution method, this is a strong hint that the kernel (Falk1) overcorrected (similiar to the artifacts from oversharpening). While undercorrection just leaves some softness, overcorrection can have more funny effects ;) The assumed PSF was too wide. With a narrower PSF (Falk2, to be developped now ;) ), the supercrop would have looked better.


Second result:

From this, I am going to conclude that the AA filter in the K-7 is a tad weaker than in the K20D. This would also easily explain the pick-up of some false color in the demosaicing process.

But let's assure that this is a very subtle effect. Just any microscopic change in focus will add the softness required to make K-7 supercrops look like K20D supercrops. Here is an example from an earlier (premature) series I shot while doing the noise tests. Now, the difference is gone.



The images are with a Pentax DA 70mm Ltd. at f/5.0. Except for the last, shot with a Zeiss 50mm at f/4.0. And I did verify that a Zeiss 50mm at f/4.0 produces the same bubble artifacts when in precise focus.

One last thing ...

The images shown so far have all been synthetic. I.e., they are downsampled from the superresolved images. Here are three unaltered sample resolution images as shot:



Unaltered resolution test shot, K-7 + DA 70mm Ltd. @ f/5.0
(Note: underexposed on purpose)


Unaltered resolution test shot, K20D + DA 70mm Ltd. @ f/5.0


Unaltered resolution test shot, K-7 + Zeiss 50mm @ f/4.0 (from "premature series", 4.5% larger)

Interestingly, the unaltered sample image from the K-7 looks just a tad softer than from the K20D. Which is why I showed a third sample (with the Zeiss) to verify that this is within a range of insignificant variation. This is the reason why I consider the matter to be intricate still. Some samples tell a different story than other samples ... But it is all in the subtle details, anyway.


Verdict:



Pentax K-7 and K20D both have an AA filter of very similiar strength. It is pretty hard to say which one is weaker. I take my bet and say the K-7 has the weaker AA filter.

But by just so little that in almost all normal shooting conditions, I expect to see no significant difference in practice. The above image summarizes the data this verdict is based upon.


These and more test images are in the following image gallery:
Resolution and Noise of K-7 vs. K20D
Please, visit for further details.



Still here?

Well, one more last thing ...

Myself (and I know of at least one other alpha tester as well) have been wondering why the images on the rear screen, when magnified after the shot, don't look as crisp as with some other makes. One could think that the images are a bit soft. So, I made a test and compared the magnified image on the rear screen with a 1:1 crop as developed by Lightroom with default settings.



As you can see, the image on the rear screen, when magnified after the shot, does indeed look a bit soft. Even if it is perfectly crisp on a computer screen. I verified that this holds true for in-camera JPGs as well (to a slightly lesser extent). I think that this is something to keep in mind when working with the K-7. Even though it has a VGA screen and goes up to 32x magnification, it never shows the pixels. Probably a feature rather than a bug;) This may have been part of early rumors that the K-7 produces softer images than the K20D. My tests have not been able to confirm this.


Still here?

Then you must be waiting for the catwalk models ;)
Thanks for your patience and interest.